~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/jbd2/transaction.c

Version: ~ [ linux-5.1-rc1 ] ~ [ linux-5.0.2 ] ~ [ linux-4.20.16 ] ~ [ linux-4.19.29 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.106 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.163 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.176 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.136 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.63 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0+
  2 /*
  3  * linux/fs/jbd2/transaction.c
  4  *
  5  * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
  6  *
  7  * Copyright 1998 Red Hat corp --- All Rights Reserved
  8  *
  9  * Generic filesystem transaction handling code; part of the ext2fs
 10  * journaling system.
 11  *
 12  * This file manages transactions (compound commits managed by the
 13  * journaling code) and handles (individual atomic operations by the
 14  * filesystem).
 15  */
 16 
 17 #include <linux/time.h>
 18 #include <linux/fs.h>
 19 #include <linux/jbd2.h>
 20 #include <linux/errno.h>
 21 #include <linux/slab.h>
 22 #include <linux/timer.h>
 23 #include <linux/mm.h>
 24 #include <linux/highmem.h>
 25 #include <linux/hrtimer.h>
 26 #include <linux/backing-dev.h>
 27 #include <linux/bug.h>
 28 #include <linux/module.h>
 29 #include <linux/sched/mm.h>
 30 
 31 #include <trace/events/jbd2.h>
 32 
 33 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh);
 34 static void __jbd2_journal_unfile_buffer(struct journal_head *jh);
 35 
 36 static struct kmem_cache *transaction_cache;
 37 int __init jbd2_journal_init_transaction_cache(void)
 38 {
 39         J_ASSERT(!transaction_cache);
 40         transaction_cache = kmem_cache_create("jbd2_transaction_s",
 41                                         sizeof(transaction_t),
 42                                         0,
 43                                         SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY,
 44                                         NULL);
 45         if (transaction_cache)
 46                 return 0;
 47         return -ENOMEM;
 48 }
 49 
 50 void jbd2_journal_destroy_transaction_cache(void)
 51 {
 52         kmem_cache_destroy(transaction_cache);
 53         transaction_cache = NULL;
 54 }
 55 
 56 void jbd2_journal_free_transaction(transaction_t *transaction)
 57 {
 58         if (unlikely(ZERO_OR_NULL_PTR(transaction)))
 59                 return;
 60         kmem_cache_free(transaction_cache, transaction);
 61 }
 62 
 63 /*
 64  * jbd2_get_transaction: obtain a new transaction_t object.
 65  *
 66  * Simply initialise a new transaction. Initialize it in
 67  * RUNNING state and add it to the current journal (which should not
 68  * have an existing running transaction: we only make a new transaction
 69  * once we have started to commit the old one).
 70  *
 71  * Preconditions:
 72  *      The journal MUST be locked.  We don't perform atomic mallocs on the
 73  *      new transaction and we can't block without protecting against other
 74  *      processes trying to touch the journal while it is in transition.
 75  *
 76  */
 77 
 78 static void jbd2_get_transaction(journal_t *journal,
 79                                 transaction_t *transaction)
 80 {
 81         transaction->t_journal = journal;
 82         transaction->t_state = T_RUNNING;
 83         transaction->t_start_time = ktime_get();
 84         transaction->t_tid = journal->j_transaction_sequence++;
 85         transaction->t_expires = jiffies + journal->j_commit_interval;
 86         spin_lock_init(&transaction->t_handle_lock);
 87         atomic_set(&transaction->t_updates, 0);
 88         atomic_set(&transaction->t_outstanding_credits,
 89                    atomic_read(&journal->j_reserved_credits));
 90         atomic_set(&transaction->t_handle_count, 0);
 91         INIT_LIST_HEAD(&transaction->t_inode_list);
 92         INIT_LIST_HEAD(&transaction->t_private_list);
 93 
 94         /* Set up the commit timer for the new transaction. */
 95         journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires);
 96         add_timer(&journal->j_commit_timer);
 97 
 98         J_ASSERT(journal->j_running_transaction == NULL);
 99         journal->j_running_transaction = transaction;
100         transaction->t_max_wait = 0;
101         transaction->t_start = jiffies;
102         transaction->t_requested = 0;
103 }
104 
105 /*
106  * Handle management.
107  *
108  * A handle_t is an object which represents a single atomic update to a
109  * filesystem, and which tracks all of the modifications which form part
110  * of that one update.
111  */
112 
113 /*
114  * Update transaction's maximum wait time, if debugging is enabled.
115  *
116  * In order for t_max_wait to be reliable, it must be protected by a
117  * lock.  But doing so will mean that start_this_handle() can not be
118  * run in parallel on SMP systems, which limits our scalability.  So
119  * unless debugging is enabled, we no longer update t_max_wait, which
120  * means that maximum wait time reported by the jbd2_run_stats
121  * tracepoint will always be zero.
122  */
123 static inline void update_t_max_wait(transaction_t *transaction,
124                                      unsigned long ts)
125 {
126 #ifdef CONFIG_JBD2_DEBUG
127         if (jbd2_journal_enable_debug &&
128             time_after(transaction->t_start, ts)) {
129                 ts = jbd2_time_diff(ts, transaction->t_start);
130                 spin_lock(&transaction->t_handle_lock);
131                 if (ts > transaction->t_max_wait)
132                         transaction->t_max_wait = ts;
133                 spin_unlock(&transaction->t_handle_lock);
134         }
135 #endif
136 }
137 
138 /*
139  * Wait until running transaction passes to T_FLUSH state and new transaction
140  * can thus be started. Also starts the commit if needed. The function expects
141  * running transaction to exist and releases j_state_lock.
142  */
143 static void wait_transaction_locked(journal_t *journal)
144         __releases(journal->j_state_lock)
145 {
146         DEFINE_WAIT(wait);
147         int need_to_start;
148         tid_t tid = journal->j_running_transaction->t_tid;
149 
150         prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
151                         TASK_UNINTERRUPTIBLE);
152         need_to_start = !tid_geq(journal->j_commit_request, tid);
153         read_unlock(&journal->j_state_lock);
154         if (need_to_start)
155                 jbd2_log_start_commit(journal, tid);
156         jbd2_might_wait_for_commit(journal);
157         schedule();
158         finish_wait(&journal->j_wait_transaction_locked, &wait);
159 }
160 
161 /*
162  * Wait until running transaction transitions from T_SWITCH to T_FLUSH
163  * state and new transaction can thus be started. The function releases
164  * j_state_lock.
165  */
166 static void wait_transaction_switching(journal_t *journal)
167         __releases(journal->j_state_lock)
168 {
169         DEFINE_WAIT(wait);
170 
171         if (WARN_ON(!journal->j_running_transaction ||
172                     journal->j_running_transaction->t_state != T_SWITCH))
173                 return;
174         prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
175                         TASK_UNINTERRUPTIBLE);
176         read_unlock(&journal->j_state_lock);
177         /*
178          * We don't call jbd2_might_wait_for_commit() here as there's no
179          * waiting for outstanding handles happening anymore in T_SWITCH state
180          * and handling of reserved handles actually relies on that for
181          * correctness.
182          */
183         schedule();
184         finish_wait(&journal->j_wait_transaction_locked, &wait);
185 }
186 
187 static void sub_reserved_credits(journal_t *journal, int blocks)
188 {
189         atomic_sub(blocks, &journal->j_reserved_credits);
190         wake_up(&journal->j_wait_reserved);
191 }
192 
193 /*
194  * Wait until we can add credits for handle to the running transaction.  Called
195  * with j_state_lock held for reading. Returns 0 if handle joined the running
196  * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
197  * caller must retry.
198  */
199 static int add_transaction_credits(journal_t *journal, int blocks,
200                                    int rsv_blocks)
201 {
202         transaction_t *t = journal->j_running_transaction;
203         int needed;
204         int total = blocks + rsv_blocks;
205 
206         /*
207          * If the current transaction is locked down for commit, wait
208          * for the lock to be released.
209          */
210         if (t->t_state != T_RUNNING) {
211                 WARN_ON_ONCE(t->t_state >= T_FLUSH);
212                 wait_transaction_locked(journal);
213                 return 1;
214         }
215 
216         /*
217          * If there is not enough space left in the log to write all
218          * potential buffers requested by this operation, we need to
219          * stall pending a log checkpoint to free some more log space.
220          */
221         needed = atomic_add_return(total, &t->t_outstanding_credits);
222         if (needed > journal->j_max_transaction_buffers) {
223                 /*
224                  * If the current transaction is already too large,
225                  * then start to commit it: we can then go back and
226                  * attach this handle to a new transaction.
227                  */
228                 atomic_sub(total, &t->t_outstanding_credits);
229 
230                 /*
231                  * Is the number of reserved credits in the current transaction too
232                  * big to fit this handle? Wait until reserved credits are freed.
233                  */
234                 if (atomic_read(&journal->j_reserved_credits) + total >
235                     journal->j_max_transaction_buffers) {
236                         read_unlock(&journal->j_state_lock);
237                         jbd2_might_wait_for_commit(journal);
238                         wait_event(journal->j_wait_reserved,
239                                    atomic_read(&journal->j_reserved_credits) + total <=
240                                    journal->j_max_transaction_buffers);
241                         return 1;
242                 }
243 
244                 wait_transaction_locked(journal);
245                 return 1;
246         }
247 
248         /*
249          * The commit code assumes that it can get enough log space
250          * without forcing a checkpoint.  This is *critical* for
251          * correctness: a checkpoint of a buffer which is also
252          * associated with a committing transaction creates a deadlock,
253          * so commit simply cannot force through checkpoints.
254          *
255          * We must therefore ensure the necessary space in the journal
256          * *before* starting to dirty potentially checkpointed buffers
257          * in the new transaction.
258          */
259         if (jbd2_log_space_left(journal) < jbd2_space_needed(journal)) {
260                 atomic_sub(total, &t->t_outstanding_credits);
261                 read_unlock(&journal->j_state_lock);
262                 jbd2_might_wait_for_commit(journal);
263                 write_lock(&journal->j_state_lock);
264                 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal))
265                         __jbd2_log_wait_for_space(journal);
266                 write_unlock(&journal->j_state_lock);
267                 return 1;
268         }
269 
270         /* No reservation? We are done... */
271         if (!rsv_blocks)
272                 return 0;
273 
274         needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
275         /* We allow at most half of a transaction to be reserved */
276         if (needed > journal->j_max_transaction_buffers / 2) {
277                 sub_reserved_credits(journal, rsv_blocks);
278                 atomic_sub(total, &t->t_outstanding_credits);
279                 read_unlock(&journal->j_state_lock);
280                 jbd2_might_wait_for_commit(journal);
281                 wait_event(journal->j_wait_reserved,
282                          atomic_read(&journal->j_reserved_credits) + rsv_blocks
283                          <= journal->j_max_transaction_buffers / 2);
284                 return 1;
285         }
286         return 0;
287 }
288 
289 /*
290  * start_this_handle: Given a handle, deal with any locking or stalling
291  * needed to make sure that there is enough journal space for the handle
292  * to begin.  Attach the handle to a transaction and set up the
293  * transaction's buffer credits.
294  */
295 
296 static int start_this_handle(journal_t *journal, handle_t *handle,
297                              gfp_t gfp_mask)
298 {
299         transaction_t   *transaction, *new_transaction = NULL;
300         int             blocks = handle->h_buffer_credits;
301         int             rsv_blocks = 0;
302         unsigned long ts = jiffies;
303 
304         if (handle->h_rsv_handle)
305                 rsv_blocks = handle->h_rsv_handle->h_buffer_credits;
306 
307         /*
308          * Limit the number of reserved credits to 1/2 of maximum transaction
309          * size and limit the number of total credits to not exceed maximum
310          * transaction size per operation.
311          */
312         if ((rsv_blocks > journal->j_max_transaction_buffers / 2) ||
313             (rsv_blocks + blocks > journal->j_max_transaction_buffers)) {
314                 printk(KERN_ERR "JBD2: %s wants too many credits "
315                        "credits:%d rsv_credits:%d max:%d\n",
316                        current->comm, blocks, rsv_blocks,
317                        journal->j_max_transaction_buffers);
318                 WARN_ON(1);
319                 return -ENOSPC;
320         }
321 
322 alloc_transaction:
323         if (!journal->j_running_transaction) {
324                 /*
325                  * If __GFP_FS is not present, then we may be being called from
326                  * inside the fs writeback layer, so we MUST NOT fail.
327                  */
328                 if ((gfp_mask & __GFP_FS) == 0)
329                         gfp_mask |= __GFP_NOFAIL;
330                 new_transaction = kmem_cache_zalloc(transaction_cache,
331                                                     gfp_mask);
332                 if (!new_transaction)
333                         return -ENOMEM;
334         }
335 
336         jbd_debug(3, "New handle %p going live.\n", handle);
337 
338         /*
339          * We need to hold j_state_lock until t_updates has been incremented,
340          * for proper journal barrier handling
341          */
342 repeat:
343         read_lock(&journal->j_state_lock);
344         BUG_ON(journal->j_flags & JBD2_UNMOUNT);
345         if (is_journal_aborted(journal) ||
346             (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
347                 read_unlock(&journal->j_state_lock);
348                 jbd2_journal_free_transaction(new_transaction);
349                 return -EROFS;
350         }
351 
352         /*
353          * Wait on the journal's transaction barrier if necessary. Specifically
354          * we allow reserved handles to proceed because otherwise commit could
355          * deadlock on page writeback not being able to complete.
356          */
357         if (!handle->h_reserved && journal->j_barrier_count) {
358                 read_unlock(&journal->j_state_lock);
359                 wait_event(journal->j_wait_transaction_locked,
360                                 journal->j_barrier_count == 0);
361                 goto repeat;
362         }
363 
364         if (!journal->j_running_transaction) {
365                 read_unlock(&journal->j_state_lock);
366                 if (!new_transaction)
367                         goto alloc_transaction;
368                 write_lock(&journal->j_state_lock);
369                 if (!journal->j_running_transaction &&
370                     (handle->h_reserved || !journal->j_barrier_count)) {
371                         jbd2_get_transaction(journal, new_transaction);
372                         new_transaction = NULL;
373                 }
374                 write_unlock(&journal->j_state_lock);
375                 goto repeat;
376         }
377 
378         transaction = journal->j_running_transaction;
379 
380         if (!handle->h_reserved) {
381                 /* We may have dropped j_state_lock - restart in that case */
382                 if (add_transaction_credits(journal, blocks, rsv_blocks))
383                         goto repeat;
384         } else {
385                 /*
386                  * We have handle reserved so we are allowed to join T_LOCKED
387                  * transaction and we don't have to check for transaction size
388                  * and journal space. But we still have to wait while running
389                  * transaction is being switched to a committing one as it
390                  * won't wait for any handles anymore.
391                  */
392                 if (transaction->t_state == T_SWITCH) {
393                         wait_transaction_switching(journal);
394                         goto repeat;
395                 }
396                 sub_reserved_credits(journal, blocks);
397                 handle->h_reserved = 0;
398         }
399 
400         /* OK, account for the buffers that this operation expects to
401          * use and add the handle to the running transaction. 
402          */
403         update_t_max_wait(transaction, ts);
404         handle->h_transaction = transaction;
405         handle->h_requested_credits = blocks;
406         handle->h_start_jiffies = jiffies;
407         atomic_inc(&transaction->t_updates);
408         atomic_inc(&transaction->t_handle_count);
409         jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
410                   handle, blocks,
411                   atomic_read(&transaction->t_outstanding_credits),
412                   jbd2_log_space_left(journal));
413         read_unlock(&journal->j_state_lock);
414         current->journal_info = handle;
415 
416         rwsem_acquire_read(&journal->j_trans_commit_map, 0, 0, _THIS_IP_);
417         jbd2_journal_free_transaction(new_transaction);
418         /*
419          * Ensure that no allocations done while the transaction is open are
420          * going to recurse back to the fs layer.
421          */
422         handle->saved_alloc_context = memalloc_nofs_save();
423         return 0;
424 }
425 
426 /* Allocate a new handle.  This should probably be in a slab... */
427 static handle_t *new_handle(int nblocks)
428 {
429         handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
430         if (!handle)
431                 return NULL;
432         handle->h_buffer_credits = nblocks;
433         handle->h_ref = 1;
434 
435         return handle;
436 }
437 
438 handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
439                               gfp_t gfp_mask, unsigned int type,
440                               unsigned int line_no)
441 {
442         handle_t *handle = journal_current_handle();
443         int err;
444 
445         if (!journal)
446                 return ERR_PTR(-EROFS);
447 
448         if (handle) {
449                 J_ASSERT(handle->h_transaction->t_journal == journal);
450                 handle->h_ref++;
451                 return handle;
452         }
453 
454         handle = new_handle(nblocks);
455         if (!handle)
456                 return ERR_PTR(-ENOMEM);
457         if (rsv_blocks) {
458                 handle_t *rsv_handle;
459 
460                 rsv_handle = new_handle(rsv_blocks);
461                 if (!rsv_handle) {
462                         jbd2_free_handle(handle);
463                         return ERR_PTR(-ENOMEM);
464                 }
465                 rsv_handle->h_reserved = 1;
466                 rsv_handle->h_journal = journal;
467                 handle->h_rsv_handle = rsv_handle;
468         }
469 
470         err = start_this_handle(journal, handle, gfp_mask);
471         if (err < 0) {
472                 if (handle->h_rsv_handle)
473                         jbd2_free_handle(handle->h_rsv_handle);
474                 jbd2_free_handle(handle);
475                 return ERR_PTR(err);
476         }
477         handle->h_type = type;
478         handle->h_line_no = line_no;
479         trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
480                                 handle->h_transaction->t_tid, type,
481                                 line_no, nblocks);
482 
483         return handle;
484 }
485 EXPORT_SYMBOL(jbd2__journal_start);
486 
487 
488 /**
489  * handle_t *jbd2_journal_start() - Obtain a new handle.
490  * @journal: Journal to start transaction on.
491  * @nblocks: number of block buffer we might modify
492  *
493  * We make sure that the transaction can guarantee at least nblocks of
494  * modified buffers in the log.  We block until the log can guarantee
495  * that much space. Additionally, if rsv_blocks > 0, we also create another
496  * handle with rsv_blocks reserved blocks in the journal. This handle is
497  * is stored in h_rsv_handle. It is not attached to any particular transaction
498  * and thus doesn't block transaction commit. If the caller uses this reserved
499  * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
500  * on the parent handle will dispose the reserved one. Reserved handle has to
501  * be converted to a normal handle using jbd2_journal_start_reserved() before
502  * it can be used.
503  *
504  * Return a pointer to a newly allocated handle, or an ERR_PTR() value
505  * on failure.
506  */
507 handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
508 {
509         return jbd2__journal_start(journal, nblocks, 0, GFP_NOFS, 0, 0);
510 }
511 EXPORT_SYMBOL(jbd2_journal_start);
512 
513 void jbd2_journal_free_reserved(handle_t *handle)
514 {
515         journal_t *journal = handle->h_journal;
516 
517         WARN_ON(!handle->h_reserved);
518         sub_reserved_credits(journal, handle->h_buffer_credits);
519         jbd2_free_handle(handle);
520 }
521 EXPORT_SYMBOL(jbd2_journal_free_reserved);
522 
523 /**
524  * int jbd2_journal_start_reserved() - start reserved handle
525  * @handle: handle to start
526  * @type: for handle statistics
527  * @line_no: for handle statistics
528  *
529  * Start handle that has been previously reserved with jbd2_journal_reserve().
530  * This attaches @handle to the running transaction (or creates one if there's
531  * not transaction running). Unlike jbd2_journal_start() this function cannot
532  * block on journal commit, checkpointing, or similar stuff. It can block on
533  * memory allocation or frozen journal though.
534  *
535  * Return 0 on success, non-zero on error - handle is freed in that case.
536  */
537 int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
538                                 unsigned int line_no)
539 {
540         journal_t *journal = handle->h_journal;
541         int ret = -EIO;
542 
543         if (WARN_ON(!handle->h_reserved)) {
544                 /* Someone passed in normal handle? Just stop it. */
545                 jbd2_journal_stop(handle);
546                 return ret;
547         }
548         /*
549          * Usefulness of mixing of reserved and unreserved handles is
550          * questionable. So far nobody seems to need it so just error out.
551          */
552         if (WARN_ON(current->journal_info)) {
553                 jbd2_journal_free_reserved(handle);
554                 return ret;
555         }
556 
557         handle->h_journal = NULL;
558         /*
559          * GFP_NOFS is here because callers are likely from writeback or
560          * similarly constrained call sites
561          */
562         ret = start_this_handle(journal, handle, GFP_NOFS);
563         if (ret < 0) {
564                 handle->h_journal = journal;
565                 jbd2_journal_free_reserved(handle);
566                 return ret;
567         }
568         handle->h_type = type;
569         handle->h_line_no = line_no;
570         return 0;
571 }
572 EXPORT_SYMBOL(jbd2_journal_start_reserved);
573 
574 /**
575  * int jbd2_journal_extend() - extend buffer credits.
576  * @handle:  handle to 'extend'
577  * @nblocks: nr blocks to try to extend by.
578  *
579  * Some transactions, such as large extends and truncates, can be done
580  * atomically all at once or in several stages.  The operation requests
581  * a credit for a number of buffer modifications in advance, but can
582  * extend its credit if it needs more.
583  *
584  * jbd2_journal_extend tries to give the running handle more buffer credits.
585  * It does not guarantee that allocation - this is a best-effort only.
586  * The calling process MUST be able to deal cleanly with a failure to
587  * extend here.
588  *
589  * Return 0 on success, non-zero on failure.
590  *
591  * return code < 0 implies an error
592  * return code > 0 implies normal transaction-full status.
593  */
594 int jbd2_journal_extend(handle_t *handle, int nblocks)
595 {
596         transaction_t *transaction = handle->h_transaction;
597         journal_t *journal;
598         int result;
599         int wanted;
600 
601         if (is_handle_aborted(handle))
602                 return -EROFS;
603         journal = transaction->t_journal;
604 
605         result = 1;
606 
607         read_lock(&journal->j_state_lock);
608 
609         /* Don't extend a locked-down transaction! */
610         if (transaction->t_state != T_RUNNING) {
611                 jbd_debug(3, "denied handle %p %d blocks: "
612                           "transaction not running\n", handle, nblocks);
613                 goto error_out;
614         }
615 
616         spin_lock(&transaction->t_handle_lock);
617         wanted = atomic_add_return(nblocks,
618                                    &transaction->t_outstanding_credits);
619 
620         if (wanted > journal->j_max_transaction_buffers) {
621                 jbd_debug(3, "denied handle %p %d blocks: "
622                           "transaction too large\n", handle, nblocks);
623                 atomic_sub(nblocks, &transaction->t_outstanding_credits);
624                 goto unlock;
625         }
626 
627         if (wanted + (wanted >> JBD2_CONTROL_BLOCKS_SHIFT) >
628             jbd2_log_space_left(journal)) {
629                 jbd_debug(3, "denied handle %p %d blocks: "
630                           "insufficient log space\n", handle, nblocks);
631                 atomic_sub(nblocks, &transaction->t_outstanding_credits);
632                 goto unlock;
633         }
634 
635         trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
636                                  transaction->t_tid,
637                                  handle->h_type, handle->h_line_no,
638                                  handle->h_buffer_credits,
639                                  nblocks);
640 
641         handle->h_buffer_credits += nblocks;
642         handle->h_requested_credits += nblocks;
643         result = 0;
644 
645         jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
646 unlock:
647         spin_unlock(&transaction->t_handle_lock);
648 error_out:
649         read_unlock(&journal->j_state_lock);
650         return result;
651 }
652 
653 
654 /**
655  * int jbd2_journal_restart() - restart a handle .
656  * @handle:  handle to restart
657  * @nblocks: nr credits requested
658  * @gfp_mask: memory allocation flags (for start_this_handle)
659  *
660  * Restart a handle for a multi-transaction filesystem
661  * operation.
662  *
663  * If the jbd2_journal_extend() call above fails to grant new buffer credits
664  * to a running handle, a call to jbd2_journal_restart will commit the
665  * handle's transaction so far and reattach the handle to a new
666  * transaction capable of guaranteeing the requested number of
667  * credits. We preserve reserved handle if there's any attached to the
668  * passed in handle.
669  */
670 int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask)
671 {
672         transaction_t *transaction = handle->h_transaction;
673         journal_t *journal;
674         tid_t           tid;
675         int             need_to_start, ret;
676 
677         /* If we've had an abort of any type, don't even think about
678          * actually doing the restart! */
679         if (is_handle_aborted(handle))
680                 return 0;
681         journal = transaction->t_journal;
682 
683         /*
684          * First unlink the handle from its current transaction, and start the
685          * commit on that.
686          */
687         J_ASSERT(atomic_read(&transaction->t_updates) > 0);
688         J_ASSERT(journal_current_handle() == handle);
689 
690         read_lock(&journal->j_state_lock);
691         spin_lock(&transaction->t_handle_lock);
692         atomic_sub(handle->h_buffer_credits,
693                    &transaction->t_outstanding_credits);
694         if (handle->h_rsv_handle) {
695                 sub_reserved_credits(journal,
696                                      handle->h_rsv_handle->h_buffer_credits);
697         }
698         if (atomic_dec_and_test(&transaction->t_updates))
699                 wake_up(&journal->j_wait_updates);
700         tid = transaction->t_tid;
701         spin_unlock(&transaction->t_handle_lock);
702         handle->h_transaction = NULL;
703         current->journal_info = NULL;
704 
705         jbd_debug(2, "restarting handle %p\n", handle);
706         need_to_start = !tid_geq(journal->j_commit_request, tid);
707         read_unlock(&journal->j_state_lock);
708         if (need_to_start)
709                 jbd2_log_start_commit(journal, tid);
710 
711         rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_);
712         handle->h_buffer_credits = nblocks;
713         /*
714          * Restore the original nofs context because the journal restart
715          * is basically the same thing as journal stop and start.
716          * start_this_handle will start a new nofs context.
717          */
718         memalloc_nofs_restore(handle->saved_alloc_context);
719         ret = start_this_handle(journal, handle, gfp_mask);
720         return ret;
721 }
722 EXPORT_SYMBOL(jbd2__journal_restart);
723 
724 
725 int jbd2_journal_restart(handle_t *handle, int nblocks)
726 {
727         return jbd2__journal_restart(handle, nblocks, GFP_NOFS);
728 }
729 EXPORT_SYMBOL(jbd2_journal_restart);
730 
731 /**
732  * void jbd2_journal_lock_updates () - establish a transaction barrier.
733  * @journal:  Journal to establish a barrier on.
734  *
735  * This locks out any further updates from being started, and blocks
736  * until all existing updates have completed, returning only once the
737  * journal is in a quiescent state with no updates running.
738  *
739  * The journal lock should not be held on entry.
740  */
741 void jbd2_journal_lock_updates(journal_t *journal)
742 {
743         DEFINE_WAIT(wait);
744 
745         jbd2_might_wait_for_commit(journal);
746 
747         write_lock(&journal->j_state_lock);
748         ++journal->j_barrier_count;
749 
750         /* Wait until there are no reserved handles */
751         if (atomic_read(&journal->j_reserved_credits)) {
752                 write_unlock(&journal->j_state_lock);
753                 wait_event(journal->j_wait_reserved,
754                            atomic_read(&journal->j_reserved_credits) == 0);
755                 write_lock(&journal->j_state_lock);
756         }
757 
758         /* Wait until there are no running updates */
759         while (1) {
760                 transaction_t *transaction = journal->j_running_transaction;
761 
762                 if (!transaction)
763                         break;
764 
765                 spin_lock(&transaction->t_handle_lock);
766                 prepare_to_wait(&journal->j_wait_updates, &wait,
767                                 TASK_UNINTERRUPTIBLE);
768                 if (!atomic_read(&transaction->t_updates)) {
769                         spin_unlock(&transaction->t_handle_lock);
770                         finish_wait(&journal->j_wait_updates, &wait);
771                         break;
772                 }
773                 spin_unlock(&transaction->t_handle_lock);
774                 write_unlock(&journal->j_state_lock);
775                 schedule();
776                 finish_wait(&journal->j_wait_updates, &wait);
777                 write_lock(&journal->j_state_lock);
778         }
779         write_unlock(&journal->j_state_lock);
780 
781         /*
782          * We have now established a barrier against other normal updates, but
783          * we also need to barrier against other jbd2_journal_lock_updates() calls
784          * to make sure that we serialise special journal-locked operations
785          * too.
786          */
787         mutex_lock(&journal->j_barrier);
788 }
789 
790 /**
791  * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
792  * @journal:  Journal to release the barrier on.
793  *
794  * Release a transaction barrier obtained with jbd2_journal_lock_updates().
795  *
796  * Should be called without the journal lock held.
797  */
798 void jbd2_journal_unlock_updates (journal_t *journal)
799 {
800         J_ASSERT(journal->j_barrier_count != 0);
801 
802         mutex_unlock(&journal->j_barrier);
803         write_lock(&journal->j_state_lock);
804         --journal->j_barrier_count;
805         write_unlock(&journal->j_state_lock);
806         wake_up(&journal->j_wait_transaction_locked);
807 }
808 
809 static void warn_dirty_buffer(struct buffer_head *bh)
810 {
811         printk(KERN_WARNING
812                "JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). "
813                "There's a risk of filesystem corruption in case of system "
814                "crash.\n",
815                bh->b_bdev, (unsigned long long)bh->b_blocknr);
816 }
817 
818 /* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
819 static void jbd2_freeze_jh_data(struct journal_head *jh)
820 {
821         struct page *page;
822         int offset;
823         char *source;
824         struct buffer_head *bh = jh2bh(jh);
825 
826         J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n");
827         page = bh->b_page;
828         offset = offset_in_page(bh->b_data);
829         source = kmap_atomic(page);
830         /* Fire data frozen trigger just before we copy the data */
831         jbd2_buffer_frozen_trigger(jh, source + offset, jh->b_triggers);
832         memcpy(jh->b_frozen_data, source + offset, bh->b_size);
833         kunmap_atomic(source);
834 
835         /*
836          * Now that the frozen data is saved off, we need to store any matching
837          * triggers.
838          */
839         jh->b_frozen_triggers = jh->b_triggers;
840 }
841 
842 /*
843  * If the buffer is already part of the current transaction, then there
844  * is nothing we need to do.  If it is already part of a prior
845  * transaction which we are still committing to disk, then we need to
846  * make sure that we do not overwrite the old copy: we do copy-out to
847  * preserve the copy going to disk.  We also account the buffer against
848  * the handle's metadata buffer credits (unless the buffer is already
849  * part of the transaction, that is).
850  *
851  */
852 static int
853 do_get_write_access(handle_t *handle, struct journal_head *jh,
854                         int force_copy)
855 {
856         struct buffer_head *bh;
857         transaction_t *transaction = handle->h_transaction;
858         journal_t *journal;
859         int error;
860         char *frozen_buffer = NULL;
861         unsigned long start_lock, time_lock;
862 
863         if (is_handle_aborted(handle))
864                 return -EROFS;
865         journal = transaction->t_journal;
866 
867         jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
868 
869         JBUFFER_TRACE(jh, "entry");
870 repeat:
871         bh = jh2bh(jh);
872 
873         /* @@@ Need to check for errors here at some point. */
874 
875         start_lock = jiffies;
876         lock_buffer(bh);
877         jbd_lock_bh_state(bh);
878 
879         /* If it takes too long to lock the buffer, trace it */
880         time_lock = jbd2_time_diff(start_lock, jiffies);
881         if (time_lock > HZ/10)
882                 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
883                         jiffies_to_msecs(time_lock));
884 
885         /* We now hold the buffer lock so it is safe to query the buffer
886          * state.  Is the buffer dirty?
887          *
888          * If so, there are two possibilities.  The buffer may be
889          * non-journaled, and undergoing a quite legitimate writeback.
890          * Otherwise, it is journaled, and we don't expect dirty buffers
891          * in that state (the buffers should be marked JBD_Dirty
892          * instead.)  So either the IO is being done under our own
893          * control and this is a bug, or it's a third party IO such as
894          * dump(8) (which may leave the buffer scheduled for read ---
895          * ie. locked but not dirty) or tune2fs (which may actually have
896          * the buffer dirtied, ugh.)  */
897 
898         if (buffer_dirty(bh)) {
899                 /*
900                  * First question: is this buffer already part of the current
901                  * transaction or the existing committing transaction?
902                  */
903                 if (jh->b_transaction) {
904                         J_ASSERT_JH(jh,
905                                 jh->b_transaction == transaction ||
906                                 jh->b_transaction ==
907                                         journal->j_committing_transaction);
908                         if (jh->b_next_transaction)
909                                 J_ASSERT_JH(jh, jh->b_next_transaction ==
910                                                         transaction);
911                         warn_dirty_buffer(bh);
912                 }
913                 /*
914                  * In any case we need to clean the dirty flag and we must
915                  * do it under the buffer lock to be sure we don't race
916                  * with running write-out.
917                  */
918                 JBUFFER_TRACE(jh, "Journalling dirty buffer");
919                 clear_buffer_dirty(bh);
920                 set_buffer_jbddirty(bh);
921         }
922 
923         unlock_buffer(bh);
924 
925         error = -EROFS;
926         if (is_handle_aborted(handle)) {
927                 jbd_unlock_bh_state(bh);
928                 goto out;
929         }
930         error = 0;
931 
932         /*
933          * The buffer is already part of this transaction if b_transaction or
934          * b_next_transaction points to it
935          */
936         if (jh->b_transaction == transaction ||
937             jh->b_next_transaction == transaction)
938                 goto done;
939 
940         /*
941          * this is the first time this transaction is touching this buffer,
942          * reset the modified flag
943          */
944         jh->b_modified = 0;
945 
946         /*
947          * If the buffer is not journaled right now, we need to make sure it
948          * doesn't get written to disk before the caller actually commits the
949          * new data
950          */
951         if (!jh->b_transaction) {
952                 JBUFFER_TRACE(jh, "no transaction");
953                 J_ASSERT_JH(jh, !jh->b_next_transaction);
954                 JBUFFER_TRACE(jh, "file as BJ_Reserved");
955                 /*
956                  * Make sure all stores to jh (b_modified, b_frozen_data) are
957                  * visible before attaching it to the running transaction.
958                  * Paired with barrier in jbd2_write_access_granted()
959                  */
960                 smp_wmb();
961                 spin_lock(&journal->j_list_lock);
962                 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
963                 spin_unlock(&journal->j_list_lock);
964                 goto done;
965         }
966         /*
967          * If there is already a copy-out version of this buffer, then we don't
968          * need to make another one
969          */
970         if (jh->b_frozen_data) {
971                 JBUFFER_TRACE(jh, "has frozen data");
972                 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
973                 goto attach_next;
974         }
975 
976         JBUFFER_TRACE(jh, "owned by older transaction");
977         J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
978         J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction);
979 
980         /*
981          * There is one case we have to be very careful about.  If the
982          * committing transaction is currently writing this buffer out to disk
983          * and has NOT made a copy-out, then we cannot modify the buffer
984          * contents at all right now.  The essence of copy-out is that it is
985          * the extra copy, not the primary copy, which gets journaled.  If the
986          * primary copy is already going to disk then we cannot do copy-out
987          * here.
988          */
989         if (buffer_shadow(bh)) {
990                 JBUFFER_TRACE(jh, "on shadow: sleep");
991                 jbd_unlock_bh_state(bh);
992                 wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE);
993                 goto repeat;
994         }
995 
996         /*
997          * Only do the copy if the currently-owning transaction still needs it.
998          * If buffer isn't on BJ_Metadata list, the committing transaction is
999          * past that stage (here we use the fact that BH_Shadow is set under
1000          * bh_state lock together with refiling to BJ_Shadow list and at this
1001          * point we know the buffer doesn't have BH_Shadow set).
1002          *
1003          * Subtle point, though: if this is a get_undo_access, then we will be
1004          * relying on the frozen_data to contain the new value of the
1005          * committed_data record after the transaction, so we HAVE to force the
1006          * frozen_data copy in that case.
1007          */
1008         if (jh->b_jlist == BJ_Metadata || force_copy) {
1009                 JBUFFER_TRACE(jh, "generate frozen data");
1010                 if (!frozen_buffer) {
1011                         JBUFFER_TRACE(jh, "allocate memory for buffer");
1012                         jbd_unlock_bh_state(bh);
1013                         frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size,
1014                                                    GFP_NOFS | __GFP_NOFAIL);
1015                         goto repeat;
1016                 }
1017                 jh->b_frozen_data = frozen_buffer;
1018                 frozen_buffer = NULL;
1019                 jbd2_freeze_jh_data(jh);
1020         }
1021 attach_next:
1022         /*
1023          * Make sure all stores to jh (b_modified, b_frozen_data) are visible
1024          * before attaching it to the running transaction. Paired with barrier
1025          * in jbd2_write_access_granted()
1026          */
1027         smp_wmb();
1028         jh->b_next_transaction = transaction;
1029 
1030 done:
1031         jbd_unlock_bh_state(bh);
1032 
1033         /*
1034          * If we are about to journal a buffer, then any revoke pending on it is
1035          * no longer valid
1036          */
1037         jbd2_journal_cancel_revoke(handle, jh);
1038 
1039 out:
1040         if (unlikely(frozen_buffer))    /* It's usually NULL */
1041                 jbd2_free(frozen_buffer, bh->b_size);
1042 
1043         JBUFFER_TRACE(jh, "exit");
1044         return error;
1045 }
1046 
1047 /* Fast check whether buffer is already attached to the required transaction */
1048 static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh,
1049                                                         bool undo)
1050 {
1051         struct journal_head *jh;
1052         bool ret = false;
1053 
1054         /* Dirty buffers require special handling... */
1055         if (buffer_dirty(bh))
1056                 return false;
1057 
1058         /*
1059          * RCU protects us from dereferencing freed pages. So the checks we do
1060          * are guaranteed not to oops. However the jh slab object can get freed
1061          * & reallocated while we work with it. So we have to be careful. When
1062          * we see jh attached to the running transaction, we know it must stay
1063          * so until the transaction is committed. Thus jh won't be freed and
1064          * will be attached to the same bh while we run.  However it can
1065          * happen jh gets freed, reallocated, and attached to the transaction
1066          * just after we get pointer to it from bh. So we have to be careful
1067          * and recheck jh still belongs to our bh before we return success.
1068          */
1069         rcu_read_lock();
1070         if (!buffer_jbd(bh))
1071                 goto out;
1072         /* This should be bh2jh() but that doesn't work with inline functions */
1073         jh = READ_ONCE(bh->b_private);
1074         if (!jh)
1075                 goto out;
1076         /* For undo access buffer must have data copied */
1077         if (undo && !jh->b_committed_data)
1078                 goto out;
1079         if (jh->b_transaction != handle->h_transaction &&
1080             jh->b_next_transaction != handle->h_transaction)
1081                 goto out;
1082         /*
1083          * There are two reasons for the barrier here:
1084          * 1) Make sure to fetch b_bh after we did previous checks so that we
1085          * detect when jh went through free, realloc, attach to transaction
1086          * while we were checking. Paired with implicit barrier in that path.
1087          * 2) So that access to bh done after jbd2_write_access_granted()
1088          * doesn't get reordered and see inconsistent state of concurrent
1089          * do_get_write_access().
1090          */
1091         smp_mb();
1092         if (unlikely(jh->b_bh != bh))
1093                 goto out;
1094         ret = true;
1095 out:
1096         rcu_read_unlock();
1097         return ret;
1098 }
1099 
1100 /**
1101  * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
1102  * @handle: transaction to add buffer modifications to
1103  * @bh:     bh to be used for metadata writes
1104  *
1105  * Returns: error code or 0 on success.
1106  *
1107  * In full data journalling mode the buffer may be of type BJ_AsyncData,
1108  * because we're ``write()ing`` a buffer which is also part of a shared mapping.
1109  */
1110 
1111 int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
1112 {
1113         struct journal_head *jh;
1114         int rc;
1115 
1116         if (jbd2_write_access_granted(handle, bh, false))
1117                 return 0;
1118 
1119         jh = jbd2_journal_add_journal_head(bh);
1120         /* We do not want to get caught playing with fields which the
1121          * log thread also manipulates.  Make sure that the buffer
1122          * completes any outstanding IO before proceeding. */
1123         rc = do_get_write_access(handle, jh, 0);
1124         jbd2_journal_put_journal_head(jh);
1125         return rc;
1126 }
1127 
1128 
1129 /*
1130  * When the user wants to journal a newly created buffer_head
1131  * (ie. getblk() returned a new buffer and we are going to populate it
1132  * manually rather than reading off disk), then we need to keep the
1133  * buffer_head locked until it has been completely filled with new
1134  * data.  In this case, we should be able to make the assertion that
1135  * the bh is not already part of an existing transaction.
1136  *
1137  * The buffer should already be locked by the caller by this point.
1138  * There is no lock ranking violation: it was a newly created,
1139  * unlocked buffer beforehand. */
1140 
1141 /**
1142  * int jbd2_journal_get_create_access () - notify intent to use newly created bh
1143  * @handle: transaction to new buffer to
1144  * @bh: new buffer.
1145  *
1146  * Call this if you create a new bh.
1147  */
1148 int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
1149 {
1150         transaction_t *transaction = handle->h_transaction;
1151         journal_t *journal;
1152         struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1153         int err;
1154 
1155         jbd_debug(5, "journal_head %p\n", jh);
1156         err = -EROFS;
1157         if (is_handle_aborted(handle))
1158                 goto out;
1159         journal = transaction->t_journal;
1160         err = 0;
1161 
1162         JBUFFER_TRACE(jh, "entry");
1163         /*
1164          * The buffer may already belong to this transaction due to pre-zeroing
1165          * in the filesystem's new_block code.  It may also be on the previous,
1166          * committing transaction's lists, but it HAS to be in Forget state in
1167          * that case: the transaction must have deleted the buffer for it to be
1168          * reused here.
1169          */
1170         jbd_lock_bh_state(bh);
1171         J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1172                 jh->b_transaction == NULL ||
1173                 (jh->b_transaction == journal->j_committing_transaction &&
1174                           jh->b_jlist == BJ_Forget)));
1175 
1176         J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1177         J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1178 
1179         if (jh->b_transaction == NULL) {
1180                 /*
1181                  * Previous jbd2_journal_forget() could have left the buffer
1182                  * with jbddirty bit set because it was being committed. When
1183                  * the commit finished, we've filed the buffer for
1184                  * checkpointing and marked it dirty. Now we are reallocating
1185                  * the buffer so the transaction freeing it must have
1186                  * committed and so it's safe to clear the dirty bit.
1187                  */
1188                 clear_buffer_dirty(jh2bh(jh));
1189                 /* first access by this transaction */
1190                 jh->b_modified = 0;
1191 
1192                 JBUFFER_TRACE(jh, "file as BJ_Reserved");
1193                 spin_lock(&journal->j_list_lock);
1194                 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1195                 spin_unlock(&journal->j_list_lock);
1196         } else if (jh->b_transaction == journal->j_committing_transaction) {
1197                 /* first access by this transaction */
1198                 jh->b_modified = 0;
1199 
1200                 JBUFFER_TRACE(jh, "set next transaction");
1201                 spin_lock(&journal->j_list_lock);
1202                 jh->b_next_transaction = transaction;
1203                 spin_unlock(&journal->j_list_lock);
1204         }
1205         jbd_unlock_bh_state(bh);
1206 
1207         /*
1208          * akpm: I added this.  ext3_alloc_branch can pick up new indirect
1209          * blocks which contain freed but then revoked metadata.  We need
1210          * to cancel the revoke in case we end up freeing it yet again
1211          * and the reallocating as data - this would cause a second revoke,
1212          * which hits an assertion error.
1213          */
1214         JBUFFER_TRACE(jh, "cancelling revoke");
1215         jbd2_journal_cancel_revoke(handle, jh);
1216 out:
1217         jbd2_journal_put_journal_head(jh);
1218         return err;
1219 }
1220 
1221 /**
1222  * int jbd2_journal_get_undo_access() -  Notify intent to modify metadata with
1223  *     non-rewindable consequences
1224  * @handle: transaction
1225  * @bh: buffer to undo
1226  *
1227  * Sometimes there is a need to distinguish between metadata which has
1228  * been committed to disk and that which has not.  The ext3fs code uses
1229  * this for freeing and allocating space, we have to make sure that we
1230  * do not reuse freed space until the deallocation has been committed,
1231  * since if we overwrote that space we would make the delete
1232  * un-rewindable in case of a crash.
1233  *
1234  * To deal with that, jbd2_journal_get_undo_access requests write access to a
1235  * buffer for parts of non-rewindable operations such as delete
1236  * operations on the bitmaps.  The journaling code must keep a copy of
1237  * the buffer's contents prior to the undo_access call until such time
1238  * as we know that the buffer has definitely been committed to disk.
1239  *
1240  * We never need to know which transaction the committed data is part
1241  * of, buffers touched here are guaranteed to be dirtied later and so
1242  * will be committed to a new transaction in due course, at which point
1243  * we can discard the old committed data pointer.
1244  *
1245  * Returns error number or 0 on success.
1246  */
1247 int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
1248 {
1249         int err;
1250         struct journal_head *jh;
1251         char *committed_data = NULL;
1252 
1253         if (jbd2_write_access_granted(handle, bh, true))
1254                 return 0;
1255 
1256         jh = jbd2_journal_add_journal_head(bh);
1257         JBUFFER_TRACE(jh, "entry");
1258 
1259         /*
1260          * Do this first --- it can drop the journal lock, so we want to
1261          * make sure that obtaining the committed_data is done
1262          * atomically wrt. completion of any outstanding commits.
1263          */
1264         err = do_get_write_access(handle, jh, 1);
1265         if (err)
1266                 goto out;
1267 
1268 repeat:
1269         if (!jh->b_committed_data)
1270                 committed_data = jbd2_alloc(jh2bh(jh)->b_size,
1271                                             GFP_NOFS|__GFP_NOFAIL);
1272 
1273         jbd_lock_bh_state(bh);
1274         if (!jh->b_committed_data) {
1275                 /* Copy out the current buffer contents into the
1276                  * preserved, committed copy. */
1277                 JBUFFER_TRACE(jh, "generate b_committed data");
1278                 if (!committed_data) {
1279                         jbd_unlock_bh_state(bh);
1280                         goto repeat;
1281                 }
1282 
1283                 jh->b_committed_data = committed_data;
1284                 committed_data = NULL;
1285                 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1286         }
1287         jbd_unlock_bh_state(bh);
1288 out:
1289         jbd2_journal_put_journal_head(jh);
1290         if (unlikely(committed_data))
1291                 jbd2_free(committed_data, bh->b_size);
1292         return err;
1293 }
1294 
1295 /**
1296  * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1297  * @bh: buffer to trigger on
1298  * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1299  *
1300  * Set any triggers on this journal_head.  This is always safe, because
1301  * triggers for a committing buffer will be saved off, and triggers for
1302  * a running transaction will match the buffer in that transaction.
1303  *
1304  * Call with NULL to clear the triggers.
1305  */
1306 void jbd2_journal_set_triggers(struct buffer_head *bh,
1307                                struct jbd2_buffer_trigger_type *type)
1308 {
1309         struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
1310 
1311         if (WARN_ON(!jh))
1312                 return;
1313         jh->b_triggers = type;
1314         jbd2_journal_put_journal_head(jh);
1315 }
1316 
1317 void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
1318                                 struct jbd2_buffer_trigger_type *triggers)
1319 {
1320         struct buffer_head *bh = jh2bh(jh);
1321 
1322         if (!triggers || !triggers->t_frozen)
1323                 return;
1324 
1325         triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
1326 }
1327 
1328 void jbd2_buffer_abort_trigger(struct journal_head *jh,
1329                                struct jbd2_buffer_trigger_type *triggers)
1330 {
1331         if (!triggers || !triggers->t_abort)
1332                 return;
1333 
1334         triggers->t_abort(triggers, jh2bh(jh));
1335 }
1336 
1337 /**
1338  * int jbd2_journal_dirty_metadata() -  mark a buffer as containing dirty metadata
1339  * @handle: transaction to add buffer to.
1340  * @bh: buffer to mark
1341  *
1342  * mark dirty metadata which needs to be journaled as part of the current
1343  * transaction.
1344  *
1345  * The buffer must have previously had jbd2_journal_get_write_access()
1346  * called so that it has a valid journal_head attached to the buffer
1347  * head.
1348  *
1349  * The buffer is placed on the transaction's metadata list and is marked
1350  * as belonging to the transaction.
1351  *
1352  * Returns error number or 0 on success.
1353  *
1354  * Special care needs to be taken if the buffer already belongs to the
1355  * current committing transaction (in which case we should have frozen
1356  * data present for that commit).  In that case, we don't relink the
1357  * buffer: that only gets done when the old transaction finally
1358  * completes its commit.
1359  */
1360 int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
1361 {
1362         transaction_t *transaction = handle->h_transaction;
1363         journal_t *journal;
1364         struct journal_head *jh;
1365         int ret = 0;
1366 
1367         if (is_handle_aborted(handle))
1368                 return -EROFS;
1369         if (!buffer_jbd(bh))
1370                 return -EUCLEAN;
1371 
1372         /*
1373          * We don't grab jh reference here since the buffer must be part
1374          * of the running transaction.
1375          */
1376         jh = bh2jh(bh);
1377         jbd_debug(5, "journal_head %p\n", jh);
1378         JBUFFER_TRACE(jh, "entry");
1379 
1380         /*
1381          * This and the following assertions are unreliable since we may see jh
1382          * in inconsistent state unless we grab bh_state lock. But this is
1383          * crucial to catch bugs so let's do a reliable check until the
1384          * lockless handling is fully proven.
1385          */
1386         if (jh->b_transaction != transaction &&
1387             jh->b_next_transaction != transaction) {
1388                 jbd_lock_bh_state(bh);
1389                 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
1390                                 jh->b_next_transaction == transaction);
1391                 jbd_unlock_bh_state(bh);
1392         }
1393         if (jh->b_modified == 1) {
1394                 /* If it's in our transaction it must be in BJ_Metadata list. */
1395                 if (jh->b_transaction == transaction &&
1396                     jh->b_jlist != BJ_Metadata) {
1397                         jbd_lock_bh_state(bh);
1398                         if (jh->b_transaction == transaction &&
1399                             jh->b_jlist != BJ_Metadata)
1400                                 pr_err("JBD2: assertion failure: h_type=%u "
1401                                        "h_line_no=%u block_no=%llu jlist=%u\n",
1402                                        handle->h_type, handle->h_line_no,
1403                                        (unsigned long long) bh->b_blocknr,
1404                                        jh->b_jlist);
1405                         J_ASSERT_JH(jh, jh->b_transaction != transaction ||
1406                                         jh->b_jlist == BJ_Metadata);
1407                         jbd_unlock_bh_state(bh);
1408                 }
1409                 goto out;
1410         }
1411 
1412         journal = transaction->t_journal;
1413         jbd_lock_bh_state(bh);
1414 
1415         if (jh->b_modified == 0) {
1416                 /*
1417                  * This buffer's got modified and becoming part
1418                  * of the transaction. This needs to be done
1419                  * once a transaction -bzzz
1420                  */
1421                 if (handle->h_buffer_credits <= 0) {
1422                         ret = -ENOSPC;
1423                         goto out_unlock_bh;
1424                 }
1425                 jh->b_modified = 1;
1426                 handle->h_buffer_credits--;
1427         }
1428 
1429         /*
1430          * fastpath, to avoid expensive locking.  If this buffer is already
1431          * on the running transaction's metadata list there is nothing to do.
1432          * Nobody can take it off again because there is a handle open.
1433          * I _think_ we're OK here with SMP barriers - a mistaken decision will
1434          * result in this test being false, so we go in and take the locks.
1435          */
1436         if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1437                 JBUFFER_TRACE(jh, "fastpath");
1438                 if (unlikely(jh->b_transaction !=
1439                              journal->j_running_transaction)) {
1440                         printk(KERN_ERR "JBD2: %s: "
1441                                "jh->b_transaction (%llu, %p, %u) != "
1442                                "journal->j_running_transaction (%p, %u)\n",
1443                                journal->j_devname,
1444                                (unsigned long long) bh->b_blocknr,
1445                                jh->b_transaction,
1446                                jh->b_transaction ? jh->b_transaction->t_tid : 0,
1447                                journal->j_running_transaction,
1448                                journal->j_running_transaction ?
1449                                journal->j_running_transaction->t_tid : 0);
1450                         ret = -EINVAL;
1451                 }
1452                 goto out_unlock_bh;
1453         }
1454 
1455         set_buffer_jbddirty(bh);
1456 
1457         /*
1458          * Metadata already on the current transaction list doesn't
1459          * need to be filed.  Metadata on another transaction's list must
1460          * be committing, and will be refiled once the commit completes:
1461          * leave it alone for now.
1462          */
1463         if (jh->b_transaction != transaction) {
1464                 JBUFFER_TRACE(jh, "already on other transaction");
1465                 if (unlikely(((jh->b_transaction !=
1466                                journal->j_committing_transaction)) ||
1467                              (jh->b_next_transaction != transaction))) {
1468                         printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: "
1469                                "bad jh for block %llu: "
1470                                "transaction (%p, %u), "
1471                                "jh->b_transaction (%p, %u), "
1472                                "jh->b_next_transaction (%p, %u), jlist %u\n",
1473                                journal->j_devname,
1474                                (unsigned long long) bh->b_blocknr,
1475                                transaction, transaction->t_tid,
1476                                jh->b_transaction,
1477                                jh->b_transaction ?
1478                                jh->b_transaction->t_tid : 0,
1479                                jh->b_next_transaction,
1480                                jh->b_next_transaction ?
1481                                jh->b_next_transaction->t_tid : 0,
1482                                jh->b_jlist);
1483                         WARN_ON(1);
1484                         ret = -EINVAL;
1485                 }
1486                 /* And this case is illegal: we can't reuse another
1487                  * transaction's data buffer, ever. */
1488                 goto out_unlock_bh;
1489         }
1490 
1491         /* That test should have eliminated the following case: */
1492         J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
1493 
1494         JBUFFER_TRACE(jh, "file as BJ_Metadata");
1495         spin_lock(&journal->j_list_lock);
1496         __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
1497         spin_unlock(&journal->j_list_lock);
1498 out_unlock_bh:
1499         jbd_unlock_bh_state(bh);
1500 out:
1501         JBUFFER_TRACE(jh, "exit");
1502         return ret;
1503 }
1504 
1505 /**
1506  * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1507  * @handle: transaction handle
1508  * @bh:     bh to 'forget'
1509  *
1510  * We can only do the bforget if there are no commits pending against the
1511  * buffer.  If the buffer is dirty in the current running transaction we
1512  * can safely unlink it.
1513  *
1514  * bh may not be a journalled buffer at all - it may be a non-JBD
1515  * buffer which came off the hashtable.  Check for this.
1516  *
1517  * Decrements bh->b_count by one.
1518  *
1519  * Allow this call even if the handle has aborted --- it may be part of
1520  * the caller's cleanup after an abort.
1521  */
1522 int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh)
1523 {
1524         transaction_t *transaction = handle->h_transaction;
1525         journal_t *journal;
1526         struct journal_head *jh;
1527         int drop_reserve = 0;
1528         int err = 0;
1529         int was_modified = 0;
1530 
1531         if (is_handle_aborted(handle))
1532                 return -EROFS;
1533         journal = transaction->t_journal;
1534 
1535         BUFFER_TRACE(bh, "entry");
1536 
1537         jbd_lock_bh_state(bh);
1538 
1539         if (!buffer_jbd(bh))
1540                 goto not_jbd;
1541         jh = bh2jh(bh);
1542 
1543         /* Critical error: attempting to delete a bitmap buffer, maybe?
1544          * Don't do any jbd operations, and return an error. */
1545         if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1546                          "inconsistent data on disk")) {
1547                 err = -EIO;
1548                 goto not_jbd;
1549         }
1550 
1551         /* keep track of whether or not this transaction modified us */
1552         was_modified = jh->b_modified;
1553 
1554         /*
1555          * The buffer's going from the transaction, we must drop
1556          * all references -bzzz
1557          */
1558         jh->b_modified = 0;
1559 
1560         if (jh->b_transaction == transaction) {
1561                 J_ASSERT_JH(jh, !jh->b_frozen_data);
1562 
1563                 /* If we are forgetting a buffer which is already part
1564                  * of this transaction, then we can just drop it from
1565                  * the transaction immediately. */
1566                 clear_buffer_dirty(bh);
1567                 clear_buffer_jbddirty(bh);
1568 
1569                 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1570 
1571                 /*
1572                  * we only want to drop a reference if this transaction
1573                  * modified the buffer
1574                  */
1575                 if (was_modified)
1576                         drop_reserve = 1;
1577 
1578                 /*
1579                  * We are no longer going to journal this buffer.
1580                  * However, the commit of this transaction is still
1581                  * important to the buffer: the delete that we are now
1582                  * processing might obsolete an old log entry, so by
1583                  * committing, we can satisfy the buffer's checkpoint.
1584                  *
1585                  * So, if we have a checkpoint on the buffer, we should
1586                  * now refile the buffer on our BJ_Forget list so that
1587                  * we know to remove the checkpoint after we commit.
1588                  */
1589 
1590                 spin_lock(&journal->j_list_lock);
1591                 if (jh->b_cp_transaction) {
1592                         __jbd2_journal_temp_unlink_buffer(jh);
1593                         __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1594                 } else {
1595                         __jbd2_journal_unfile_buffer(jh);
1596                         if (!buffer_jbd(bh)) {
1597                                 spin_unlock(&journal->j_list_lock);
1598                                 goto not_jbd;
1599                         }
1600                 }
1601                 spin_unlock(&journal->j_list_lock);
1602         } else if (jh->b_transaction) {
1603                 J_ASSERT_JH(jh, (jh->b_transaction ==
1604                                  journal->j_committing_transaction));
1605                 /* However, if the buffer is still owned by a prior
1606                  * (committing) transaction, we can't drop it yet... */
1607                 JBUFFER_TRACE(jh, "belongs to older transaction");
1608                 /* ... but we CAN drop it from the new transaction through
1609                  * marking the buffer as freed and set j_next_transaction to
1610                  * the new transaction, so that not only the commit code
1611                  * knows it should clear dirty bits when it is done with the
1612                  * buffer, but also the buffer can be checkpointed only
1613                  * after the new transaction commits. */
1614 
1615                 set_buffer_freed(bh);
1616 
1617                 if (!jh->b_next_transaction) {
1618                         spin_lock(&journal->j_list_lock);
1619                         jh->b_next_transaction = transaction;
1620                         spin_unlock(&journal->j_list_lock);
1621                 } else {
1622                         J_ASSERT(jh->b_next_transaction == transaction);
1623 
1624                         /*
1625                          * only drop a reference if this transaction modified
1626                          * the buffer
1627                          */
1628                         if (was_modified)
1629                                 drop_reserve = 1;
1630                 }
1631         } else {
1632                 /*
1633                  * Finally, if the buffer is not belongs to any
1634                  * transaction, we can just drop it now if it has no
1635                  * checkpoint.
1636                  */
1637                 spin_lock(&journal->j_list_lock);
1638                 if (!jh->b_cp_transaction) {
1639                         JBUFFER_TRACE(jh, "belongs to none transaction");
1640                         spin_unlock(&journal->j_list_lock);
1641                         goto not_jbd;
1642                 }
1643 
1644                 /*
1645                  * Otherwise, if the buffer has been written to disk,
1646                  * it is safe to remove the checkpoint and drop it.
1647                  */
1648                 if (!buffer_dirty(bh)) {
1649                         __jbd2_journal_remove_checkpoint(jh);
1650                         spin_unlock(&journal->j_list_lock);
1651                         goto not_jbd;
1652                 }
1653 
1654                 /*
1655                  * The buffer is still not written to disk, we should
1656                  * attach this buffer to current transaction so that the
1657                  * buffer can be checkpointed only after the current
1658                  * transaction commits.
1659                  */
1660                 clear_buffer_dirty(bh);
1661                 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1662                 spin_unlock(&journal->j_list_lock);
1663         }
1664 
1665         jbd_unlock_bh_state(bh);
1666         __brelse(bh);
1667 drop:
1668         if (drop_reserve) {
1669                 /* no need to reserve log space for this block -bzzz */
1670                 handle->h_buffer_credits++;
1671         }
1672         return err;
1673 
1674 not_jbd:
1675         jbd_unlock_bh_state(bh);
1676         __bforget(bh);
1677         goto drop;
1678 }
1679 
1680 /**
1681  * int jbd2_journal_stop() - complete a transaction
1682  * @handle: transaction to complete.
1683  *
1684  * All done for a particular handle.
1685  *
1686  * There is not much action needed here.  We just return any remaining
1687  * buffer credits to the transaction and remove the handle.  The only
1688  * complication is that we need to start a commit operation if the
1689  * filesystem is marked for synchronous update.
1690  *
1691  * jbd2_journal_stop itself will not usually return an error, but it may
1692  * do so in unusual circumstances.  In particular, expect it to
1693  * return -EIO if a jbd2_journal_abort has been executed since the
1694  * transaction began.
1695  */
1696 int jbd2_journal_stop(handle_t *handle)
1697 {
1698         transaction_t *transaction = handle->h_transaction;
1699         journal_t *journal;
1700         int err = 0, wait_for_commit = 0;
1701         tid_t tid;
1702         pid_t pid;
1703 
1704         if (!transaction) {
1705                 /*
1706                  * Handle is already detached from the transaction so
1707                  * there is nothing to do other than decrease a refcount,
1708                  * or free the handle if refcount drops to zero
1709                  */
1710                 if (--handle->h_ref > 0) {
1711                         jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1712                                                          handle->h_ref);
1713                         return err;
1714                 } else {
1715                         if (handle->h_rsv_handle)
1716                                 jbd2_free_handle(handle->h_rsv_handle);
1717                         goto free_and_exit;
1718                 }
1719         }
1720         journal = transaction->t_journal;
1721 
1722         J_ASSERT(journal_current_handle() == handle);
1723 
1724         if (is_handle_aborted(handle))
1725                 err = -EIO;
1726         else
1727                 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
1728 
1729         if (--handle->h_ref > 0) {
1730                 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1731                           handle->h_ref);
1732                 return err;
1733         }
1734 
1735         jbd_debug(4, "Handle %p going down\n", handle);
1736         trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
1737                                 transaction->t_tid,
1738                                 handle->h_type, handle->h_line_no,
1739                                 jiffies - handle->h_start_jiffies,
1740                                 handle->h_sync, handle->h_requested_credits,
1741                                 (handle->h_requested_credits -
1742                                  handle->h_buffer_credits));
1743 
1744         /*
1745          * Implement synchronous transaction batching.  If the handle
1746          * was synchronous, don't force a commit immediately.  Let's
1747          * yield and let another thread piggyback onto this
1748          * transaction.  Keep doing that while new threads continue to
1749          * arrive.  It doesn't cost much - we're about to run a commit
1750          * and sleep on IO anyway.  Speeds up many-threaded, many-dir
1751          * operations by 30x or more...
1752          *
1753          * We try and optimize the sleep time against what the
1754          * underlying disk can do, instead of having a static sleep
1755          * time.  This is useful for the case where our storage is so
1756          * fast that it is more optimal to go ahead and force a flush
1757          * and wait for the transaction to be committed than it is to
1758          * wait for an arbitrary amount of time for new writers to
1759          * join the transaction.  We achieve this by measuring how
1760          * long it takes to commit a transaction, and compare it with
1761          * how long this transaction has been running, and if run time
1762          * < commit time then we sleep for the delta and commit.  This
1763          * greatly helps super fast disks that would see slowdowns as
1764          * more threads started doing fsyncs.
1765          *
1766          * But don't do this if this process was the most recent one
1767          * to perform a synchronous write.  We do this to detect the
1768          * case where a single process is doing a stream of sync
1769          * writes.  No point in waiting for joiners in that case.
1770          *
1771          * Setting max_batch_time to 0 disables this completely.
1772          */
1773         pid = current->pid;
1774         if (handle->h_sync && journal->j_last_sync_writer != pid &&
1775             journal->j_max_batch_time) {
1776                 u64 commit_time, trans_time;
1777 
1778                 journal->j_last_sync_writer = pid;
1779 
1780                 read_lock(&journal->j_state_lock);
1781                 commit_time = journal->j_average_commit_time;
1782                 read_unlock(&journal->j_state_lock);
1783 
1784                 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1785                                                    transaction->t_start_time));
1786 
1787                 commit_time = max_t(u64, commit_time,
1788                                     1000*journal->j_min_batch_time);
1789                 commit_time = min_t(u64, commit_time,
1790                                     1000*journal->j_max_batch_time);
1791 
1792                 if (trans_time < commit_time) {
1793                         ktime_t expires = ktime_add_ns(ktime_get(),
1794                                                        commit_time);
1795                         set_current_state(TASK_UNINTERRUPTIBLE);
1796                         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1797                 }
1798         }
1799 
1800         if (handle->h_sync)
1801                 transaction->t_synchronous_commit = 1;
1802         current->journal_info = NULL;
1803         atomic_sub(handle->h_buffer_credits,
1804                    &transaction->t_outstanding_credits);
1805 
1806         /*
1807          * If the handle is marked SYNC, we need to set another commit
1808          * going!  We also want to force a commit if the current
1809          * transaction is occupying too much of the log, or if the
1810          * transaction is too old now.
1811          */
1812         if (handle->h_sync ||
1813             (atomic_read(&transaction->t_outstanding_credits) >
1814              journal->j_max_transaction_buffers) ||
1815             time_after_eq(jiffies, transaction->t_expires)) {
1816                 /* Do this even for aborted journals: an abort still
1817                  * completes the commit thread, it just doesn't write
1818                  * anything to disk. */
1819 
1820                 jbd_debug(2, "transaction too old, requesting commit for "
1821                                         "handle %p\n", handle);
1822                 /* This is non-blocking */
1823                 jbd2_log_start_commit(journal, transaction->t_tid);
1824 
1825                 /*
1826                  * Special case: JBD2_SYNC synchronous updates require us
1827                  * to wait for the commit to complete.
1828                  */
1829                 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
1830                         wait_for_commit = 1;
1831         }
1832 
1833         /*
1834          * Once we drop t_updates, if it goes to zero the transaction
1835          * could start committing on us and eventually disappear.  So
1836          * once we do this, we must not dereference transaction
1837          * pointer again.
1838          */
1839         tid = transaction->t_tid;
1840         if (atomic_dec_and_test(&transaction->t_updates)) {
1841                 wake_up(&journal->j_wait_updates);
1842                 if (journal->j_barrier_count)
1843                         wake_up(&journal->j_wait_transaction_locked);
1844         }
1845 
1846         rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_);
1847 
1848         if (wait_for_commit)
1849                 err = jbd2_log_wait_commit(journal, tid);
1850 
1851         if (handle->h_rsv_handle)
1852                 jbd2_journal_free_reserved(handle->h_rsv_handle);
1853 free_and_exit:
1854         /*
1855          * Scope of the GFP_NOFS context is over here and so we can restore the
1856          * original alloc context.
1857          */
1858         memalloc_nofs_restore(handle->saved_alloc_context);
1859         jbd2_free_handle(handle);
1860         return err;
1861 }
1862 
1863 /*
1864  *
1865  * List management code snippets: various functions for manipulating the
1866  * transaction buffer lists.
1867  *
1868  */
1869 
1870 /*
1871  * Append a buffer to a transaction list, given the transaction's list head
1872  * pointer.
1873  *
1874  * j_list_lock is held.
1875  *
1876  * jbd_lock_bh_state(jh2bh(jh)) is held.
1877  */
1878 
1879 static inline void
1880 __blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1881 {
1882         if (!*list) {
1883                 jh->b_tnext = jh->b_tprev = jh;
1884                 *list = jh;
1885         } else {
1886                 /* Insert at the tail of the list to preserve order */
1887                 struct journal_head *first = *list, *last = first->b_tprev;
1888                 jh->b_tprev = last;
1889                 jh->b_tnext = first;
1890                 last->b_tnext = first->b_tprev = jh;
1891         }
1892 }
1893 
1894 /*
1895  * Remove a buffer from a transaction list, given the transaction's list
1896  * head pointer.
1897  *
1898  * Called with j_list_lock held, and the journal may not be locked.
1899  *
1900  * jbd_lock_bh_state(jh2bh(jh)) is held.
1901  */
1902 
1903 static inline void
1904 __blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1905 {
1906         if (*list == jh) {
1907                 *list = jh->b_tnext;
1908                 if (*list == jh)
1909                         *list = NULL;
1910         }
1911         jh->b_tprev->b_tnext = jh->b_tnext;
1912         jh->b_tnext->b_tprev = jh->b_tprev;
1913 }
1914 
1915 /*
1916  * Remove a buffer from the appropriate transaction list.
1917  *
1918  * Note that this function can *change* the value of
1919  * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1920  * t_reserved_list.  If the caller is holding onto a copy of one of these
1921  * pointers, it could go bad.  Generally the caller needs to re-read the
1922  * pointer from the transaction_t.
1923  *
1924  * Called under j_list_lock.
1925  */
1926 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
1927 {
1928         struct journal_head **list = NULL;
1929         transaction_t *transaction;
1930         struct buffer_head *bh = jh2bh(jh);
1931 
1932         J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
1933         transaction = jh->b_transaction;
1934         if (transaction)
1935                 assert_spin_locked(&transaction->t_journal->j_list_lock);
1936 
1937         J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
1938         if (jh->b_jlist != BJ_None)
1939                 J_ASSERT_JH(jh, transaction != NULL);
1940 
1941         switch (jh->b_jlist) {
1942         case BJ_None:
1943                 return;
1944         case BJ_Metadata:
1945                 transaction->t_nr_buffers--;
1946                 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
1947                 list = &transaction->t_buffers;
1948                 break;
1949         case BJ_Forget:
1950                 list = &transaction->t_forget;
1951                 break;
1952         case BJ_Shadow:
1953                 list = &transaction->t_shadow_list;
1954                 break;
1955         case BJ_Reserved:
1956                 list = &transaction->t_reserved_list;
1957                 break;
1958         }
1959 
1960         __blist_del_buffer(list, jh);
1961         jh->b_jlist = BJ_None;
1962         if (transaction && is_journal_aborted(transaction->t_journal))
1963                 clear_buffer_jbddirty(bh);
1964         else if (test_clear_buffer_jbddirty(bh))
1965                 mark_buffer_dirty(bh);  /* Expose it to the VM */
1966 }
1967 
1968 /*
1969  * Remove buffer from all transactions.
1970  *
1971  * Called with bh_state lock and j_list_lock
1972  *
1973  * jh and bh may be already freed when this function returns.
1974  */
1975 static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
1976 {
1977         __jbd2_journal_temp_unlink_buffer(jh);
1978         jh->b_transaction = NULL;
1979         jbd2_journal_put_journal_head(jh);
1980 }
1981 
1982 void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
1983 {
1984         struct buffer_head *bh = jh2bh(jh);
1985 
1986         /* Get reference so that buffer cannot be freed before we unlock it */
1987         get_bh(bh);
1988         jbd_lock_bh_state(bh);
1989         spin_lock(&journal->j_list_lock);
1990         __jbd2_journal_unfile_buffer(jh);
1991         spin_unlock(&journal->j_list_lock);
1992         jbd_unlock_bh_state(bh);
1993         __brelse(bh);
1994 }
1995 
1996 /*
1997  * Called from jbd2_journal_try_to_free_buffers().
1998  *
1999  * Called under jbd_lock_bh_state(bh)
2000  */
2001 static void
2002 __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
2003 {
2004         struct journal_head *jh;
2005 
2006         jh = bh2jh(bh);
2007 
2008         if (buffer_locked(bh) || buffer_dirty(bh))
2009                 goto out;
2010 
2011         if (jh->b_next_transaction != NULL || jh->b_transaction != NULL)
2012                 goto out;
2013 
2014         spin_lock(&journal->j_list_lock);
2015         if (jh->b_cp_transaction != NULL) {
2016                 /* written-back checkpointed metadata buffer */
2017                 JBUFFER_TRACE(jh, "remove from checkpoint list");
2018                 __jbd2_journal_remove_checkpoint(jh);
2019         }
2020         spin_unlock(&journal->j_list_lock);
2021 out:
2022         return;
2023 }
2024 
2025 /**
2026  * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
2027  * @journal: journal for operation
2028  * @page: to try and free
2029  * @gfp_mask: we use the mask to detect how hard should we try to release
2030  * buffers. If __GFP_DIRECT_RECLAIM and __GFP_FS is set, we wait for commit
2031  * code to release the buffers.
2032  *
2033  *
2034  * For all the buffers on this page,
2035  * if they are fully written out ordered data, move them onto BUF_CLEAN
2036  * so try_to_free_buffers() can reap them.
2037  *
2038  * This function returns non-zero if we wish try_to_free_buffers()
2039  * to be called. We do this if the page is releasable by try_to_free_buffers().
2040  * We also do it if the page has locked or dirty buffers and the caller wants
2041  * us to perform sync or async writeout.
2042  *
2043  * This complicates JBD locking somewhat.  We aren't protected by the
2044  * BKL here.  We wish to remove the buffer from its committing or
2045  * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
2046  *
2047  * This may *change* the value of transaction_t->t_datalist, so anyone
2048  * who looks at t_datalist needs to lock against this function.
2049  *
2050  * Even worse, someone may be doing a jbd2_journal_dirty_data on this
2051  * buffer.  So we need to lock against that.  jbd2_journal_dirty_data()
2052  * will come out of the lock with the buffer dirty, which makes it
2053  * ineligible for release here.
2054  *
2055  * Who else is affected by this?  hmm...  Really the only contender
2056  * is do_get_write_access() - it could be looking at the buffer while
2057  * journal_try_to_free_buffer() is changing its state.  But that
2058  * cannot happen because we never reallocate freed data as metadata
2059  * while the data is part of a transaction.  Yes?
2060  *
2061  * Return 0 on failure, 1 on success
2062  */
2063 int jbd2_journal_try_to_free_buffers(journal_t *journal,
2064                                 struct page *page, gfp_t gfp_mask)
2065 {
2066         struct buffer_head *head;
2067         struct buffer_head *bh;
2068         int ret = 0;
2069 
2070         J_ASSERT(PageLocked(page));
2071 
2072         head = page_buffers(page);
2073         bh = head;
2074         do {
2075                 struct journal_head *jh;
2076 
2077                 /*
2078                  * We take our own ref against the journal_head here to avoid
2079                  * having to add tons of locking around each instance of
2080                  * jbd2_journal_put_journal_head().
2081                  */
2082                 jh = jbd2_journal_grab_journal_head(bh);
2083                 if (!jh)
2084                         continue;
2085 
2086                 jbd_lock_bh_state(bh);
2087                 __journal_try_to_free_buffer(journal, bh);
2088                 jbd2_journal_put_journal_head(jh);
2089                 jbd_unlock_bh_state(bh);
2090                 if (buffer_jbd(bh))
2091                         goto busy;
2092         } while ((bh = bh->b_this_page) != head);
2093 
2094         ret = try_to_free_buffers(page);
2095 
2096 busy:
2097         return ret;
2098 }
2099 
2100 /*
2101  * This buffer is no longer needed.  If it is on an older transaction's
2102  * checkpoint list we need to record it on this transaction's forget list
2103  * to pin this buffer (and hence its checkpointing transaction) down until
2104  * this transaction commits.  If the buffer isn't on a checkpoint list, we
2105  * release it.
2106  * Returns non-zero if JBD no longer has an interest in the buffer.
2107  *
2108  * Called under j_list_lock.
2109  *
2110  * Called under jbd_lock_bh_state(bh).
2111  */
2112 static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
2113 {
2114         int may_free = 1;
2115         struct buffer_head *bh = jh2bh(jh);
2116 
2117         if (jh->b_cp_transaction) {
2118                 JBUFFER_TRACE(jh, "on running+cp transaction");
2119                 __jbd2_journal_temp_unlink_buffer(jh);
2120                 /*
2121                  * We don't want to write the buffer anymore, clear the
2122                  * bit so that we don't confuse checks in
2123                  * __journal_file_buffer
2124                  */
2125                 clear_buffer_dirty(bh);
2126                 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
2127                 may_free = 0;
2128         } else {
2129                 JBUFFER_TRACE(jh, "on running transaction");
2130                 __jbd2_journal_unfile_buffer(jh);
2131         }
2132         return may_free;
2133 }
2134 
2135 /*
2136  * jbd2_journal_invalidatepage
2137  *
2138  * This code is tricky.  It has a number of cases to deal with.
2139  *
2140  * There are two invariants which this code relies on:
2141  *
2142  * i_size must be updated on disk before we start calling invalidatepage on the
2143  * data.
2144  *
2145  *  This is done in ext3 by defining an ext3_setattr method which
2146  *  updates i_size before truncate gets going.  By maintaining this
2147  *  invariant, we can be sure that it is safe to throw away any buffers
2148  *  attached to the current transaction: once the transaction commits,
2149  *  we know that the data will not be needed.
2150  *
2151  *  Note however that we can *not* throw away data belonging to the
2152  *  previous, committing transaction!
2153  *
2154  * Any disk blocks which *are* part of the previous, committing
2155  * transaction (and which therefore cannot be discarded immediately) are
2156  * not going to be reused in the new running transaction
2157  *
2158  *  The bitmap committed_data images guarantee this: any block which is
2159  *  allocated in one transaction and removed in the next will be marked
2160  *  as in-use in the committed_data bitmap, so cannot be reused until
2161  *  the next transaction to delete the block commits.  This means that
2162  *  leaving committing buffers dirty is quite safe: the disk blocks
2163  *  cannot be reallocated to a different file and so buffer aliasing is
2164  *  not possible.
2165  *
2166  *
2167  * The above applies mainly to ordered data mode.  In writeback mode we
2168  * don't make guarantees about the order in which data hits disk --- in
2169  * particular we don't guarantee that new dirty data is flushed before
2170  * transaction commit --- so it is always safe just to discard data
2171  * immediately in that mode.  --sct
2172  */
2173 
2174 /*
2175  * The journal_unmap_buffer helper function returns zero if the buffer
2176  * concerned remains pinned as an anonymous buffer belonging to an older
2177  * transaction.
2178  *
2179  * We're outside-transaction here.  Either or both of j_running_transaction
2180  * and j_committing_transaction may be NULL.
2181  */
2182 static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
2183                                 int partial_page)
2184 {
2185         transaction_t *transaction;
2186         struct journal_head *jh;
2187         int may_free = 1;
2188 
2189         BUFFER_TRACE(bh, "entry");
2190 
2191         /*
2192          * It is safe to proceed here without the j_list_lock because the
2193          * buffers cannot be stolen by try_to_free_buffers as long as we are
2194          * holding the page lock. --sct
2195          */
2196 
2197         if (!buffer_jbd(bh))
2198                 goto zap_buffer_unlocked;
2199 
2200         /* OK, we have data buffer in journaled mode */
2201         write_lock(&journal->j_state_lock);
2202         jbd_lock_bh_state(bh);
2203         spin_lock(&journal->j_list_lock);
2204 
2205         jh = jbd2_journal_grab_journal_head(bh);
2206         if (!jh)
2207                 goto zap_buffer_no_jh;
2208 
2209         /*
2210          * We cannot remove the buffer from checkpoint lists until the
2211          * transaction adding inode to orphan list (let's call it T)
2212          * is committed.  Otherwise if the transaction changing the
2213          * buffer would be cleaned from the journal before T is
2214          * committed, a crash will cause that the correct contents of
2215          * the buffer will be lost.  On the other hand we have to
2216          * clear the buffer dirty bit at latest at the moment when the
2217          * transaction marking the buffer as freed in the filesystem
2218          * structures is committed because from that moment on the
2219          * block can be reallocated and used by a different page.
2220          * Since the block hasn't been freed yet but the inode has
2221          * already been added to orphan list, it is safe for us to add
2222          * the buffer to BJ_Forget list of the newest transaction.
2223          *
2224          * Also we have to clear buffer_mapped flag of a truncated buffer
2225          * because the buffer_head may be attached to the page straddling
2226          * i_size (can happen only when blocksize < pagesize) and thus the
2227          * buffer_head can be reused when the file is extended again. So we end
2228          * up keeping around invalidated buffers attached to transactions'
2229          * BJ_Forget list just to stop checkpointing code from cleaning up
2230          * the transaction this buffer was modified in.
2231          */
2232         transaction = jh->b_transaction;
2233         if (transaction == NULL) {
2234                 /* First case: not on any transaction.  If it
2235                  * has no checkpoint link, then we can zap it:
2236                  * it's a writeback-mode buffer so we don't care
2237                  * if it hits disk safely. */
2238                 if (!jh->b_cp_transaction) {
2239                         JBUFFER_TRACE(jh, "not on any transaction: zap");
2240                         goto zap_buffer;
2241                 }
2242 
2243                 if (!buffer_dirty(bh)) {
2244                         /* bdflush has written it.  We can drop it now */
2245                         __jbd2_journal_remove_checkpoint(jh);
2246                         goto zap_buffer;
2247                 }
2248 
2249                 /* OK, it must be in the journal but still not
2250                  * written fully to disk: it's metadata or
2251                  * journaled data... */
2252 
2253                 if (journal->j_running_transaction) {
2254                         /* ... and once the current transaction has
2255                          * committed, the buffer won't be needed any
2256                          * longer. */
2257                         JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
2258                         may_free = __dispose_buffer(jh,
2259                                         journal->j_running_transaction);
2260                         goto zap_buffer;
2261                 } else {
2262                         /* There is no currently-running transaction. So the
2263                          * orphan record which we wrote for this file must have
2264                          * passed into commit.  We must attach this buffer to
2265                          * the committing transaction, if it exists. */
2266                         if (journal->j_committing_transaction) {
2267                                 JBUFFER_TRACE(jh, "give to committing trans");
2268                                 may_free = __dispose_buffer(jh,
2269                                         journal->j_committing_transaction);
2270                                 goto zap_buffer;
2271                         } else {
2272                                 /* The orphan record's transaction has
2273                                  * committed.  We can cleanse this buffer */
2274                                 clear_buffer_jbddirty(bh);
2275                                 __jbd2_journal_remove_checkpoint(jh);
2276                                 goto zap_buffer;
2277                         }
2278                 }
2279         } else if (transaction == journal->j_committing_transaction) {
2280                 JBUFFER_TRACE(jh, "on committing transaction");
2281                 /*
2282                  * The buffer is committing, we simply cannot touch
2283                  * it. If the page is straddling i_size we have to wait
2284                  * for commit and try again.
2285                  */
2286                 if (partial_page) {
2287                         jbd2_journal_put_journal_head(jh);
2288                         spin_unlock(&journal->j_list_lock);
2289                         jbd_unlock_bh_state(bh);
2290                         write_unlock(&journal->j_state_lock);
2291                         return -EBUSY;
2292                 }
2293                 /*
2294                  * OK, buffer won't be reachable after truncate. We just set
2295                  * j_next_transaction to the running transaction (if there is
2296                  * one) and mark buffer as freed so that commit code knows it
2297                  * should clear dirty bits when it is done with the buffer.
2298                  */
2299                 set_buffer_freed(bh);
2300                 if (journal->j_running_transaction && buffer_jbddirty(bh))
2301                         jh->b_next_transaction = journal->j_running_transaction;
2302                 jbd2_journal_put_journal_head(jh);
2303                 spin_unlock(&journal->j_list_lock);
2304                 jbd_unlock_bh_state(bh);
2305                 write_unlock(&journal->j_state_lock);
2306                 return 0;
2307         } else {
2308                 /* Good, the buffer belongs to the running transaction.
2309                  * We are writing our own transaction's data, not any
2310                  * previous one's, so it is safe to throw it away
2311                  * (remember that we expect the filesystem to have set
2312                  * i_size already for this truncate so recovery will not
2313                  * expose the disk blocks we are discarding here.) */
2314                 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
2315                 JBUFFER_TRACE(jh, "on running transaction");
2316                 may_free = __dispose_buffer(jh, transaction);
2317         }
2318 
2319 zap_buffer:
2320         /*
2321          * This is tricky. Although the buffer is truncated, it may be reused
2322          * if blocksize < pagesize and it is attached to the page straddling
2323          * EOF. Since the buffer might have been added to BJ_Forget list of the
2324          * running transaction, journal_get_write_access() won't clear
2325          * b_modified and credit accounting gets confused. So clear b_modified
2326          * here.
2327          */
2328         jh->b_modified = 0;
2329         jbd2_journal_put_journal_head(jh);
2330 zap_buffer_no_jh:
2331         spin_unlock(&journal->j_list_lock);
2332         jbd_unlock_bh_state(bh);
2333         write_unlock(&journal->j_state_lock);
2334 zap_buffer_unlocked:
2335         clear_buffer_dirty(bh);
2336         J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2337         clear_buffer_mapped(bh);
2338         clear_buffer_req(bh);
2339         clear_buffer_new(bh);
2340         clear_buffer_delay(bh);
2341         clear_buffer_unwritten(bh);
2342         bh->b_bdev = NULL;
2343         return may_free;
2344 }
2345 
2346 /**
2347  * void jbd2_journal_invalidatepage()
2348  * @journal: journal to use for flush...
2349  * @page:    page to flush
2350  * @offset:  start of the range to invalidate
2351  * @length:  length of the range to invalidate
2352  *
2353  * Reap page buffers containing data after in the specified range in page.
2354  * Can return -EBUSY if buffers are part of the committing transaction and
2355  * the page is straddling i_size. Caller then has to wait for current commit
2356  * and try again.
2357  */
2358 int jbd2_journal_invalidatepage(journal_t *journal,
2359                                 struct page *page,
2360                                 unsigned int offset,
2361                                 unsigned int length)
2362 {
2363         struct buffer_head *head, *bh, *next;
2364         unsigned int stop = offset + length;
2365         unsigned int curr_off = 0;
2366         int partial_page = (offset || length < PAGE_SIZE);
2367         int may_free = 1;
2368         int ret = 0;
2369 
2370         if (!PageLocked(page))
2371                 BUG();
2372         if (!page_has_buffers(page))
2373                 return 0;
2374 
2375         BUG_ON(stop > PAGE_SIZE || stop < length);
2376 
2377         /* We will potentially be playing with lists other than just the
2378          * data lists (especially for journaled data mode), so be
2379          * cautious in our locking. */
2380 
2381         head = bh = page_buffers(page);
2382         do {
2383                 unsigned int next_off = curr_off + bh->b_size;
2384                 next = bh->b_this_page;
2385 
2386                 if (next_off > stop)
2387                         return 0;
2388 
2389                 if (offset <= curr_off) {
2390                         /* This block is wholly outside the truncation point */
2391                         lock_buffer(bh);
2392                         ret = journal_unmap_buffer(journal, bh, partial_page);
2393                         unlock_buffer(bh);
2394                         if (ret < 0)
2395                                 return ret;
2396                         may_free &= ret;
2397                 }
2398                 curr_off = next_off;
2399                 bh = next;
2400 
2401         } while (bh != head);
2402 
2403         if (!partial_page) {
2404                 if (may_free && try_to_free_buffers(page))
2405                         J_ASSERT(!page_has_buffers(page));
2406         }
2407         return 0;
2408 }
2409 
2410 /*
2411  * File a buffer on the given transaction list.
2412  */
2413 void __jbd2_journal_file_buffer(struct journal_head *jh,
2414                         transaction_t *transaction, int jlist)
2415 {
2416         struct journal_head **list = NULL;
2417         int was_dirty = 0;
2418         struct buffer_head *bh = jh2bh(jh);
2419 
2420         J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2421         assert_spin_locked(&transaction->t_journal->j_list_lock);
2422 
2423         J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2424         J_ASSERT_JH(jh, jh->b_transaction == transaction ||
2425                                 jh->b_transaction == NULL);
2426 
2427         if (jh->b_transaction && jh->b_jlist == jlist)
2428                 return;
2429 
2430         if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2431             jlist == BJ_Shadow || jlist == BJ_Forget) {
2432                 /*
2433                  * For metadata buffers, we track dirty bit in buffer_jbddirty
2434                  * instead of buffer_dirty. We should not see a dirty bit set
2435                  * here because we clear it in do_get_write_access but e.g.
2436                  * tune2fs can modify the sb and set the dirty bit at any time
2437                  * so we try to gracefully handle that.
2438                  */
2439                 if (buffer_dirty(bh))
2440                         warn_dirty_buffer(bh);
2441                 if (test_clear_buffer_dirty(bh) ||
2442                     test_clear_buffer_jbddirty(bh))
2443                         was_dirty = 1;
2444         }
2445 
2446         if (jh->b_transaction)
2447                 __jbd2_journal_temp_unlink_buffer(jh);
2448         else
2449                 jbd2_journal_grab_journal_head(bh);
2450         jh->b_transaction = transaction;
2451 
2452         switch (jlist) {
2453         case BJ_None:
2454                 J_ASSERT_JH(jh, !jh->b_committed_data);
2455                 J_ASSERT_JH(jh, !jh->b_frozen_data);
2456                 return;
2457         case BJ_Metadata:
2458                 transaction->t_nr_buffers++;
2459                 list = &transaction->t_buffers;
2460                 break;
2461         case BJ_Forget:
2462                 list = &transaction->t_forget;
2463                 break;
2464         case BJ_Shadow:
2465                 list = &transaction->t_shadow_list;
2466                 break;
2467         case BJ_Reserved:
2468                 list = &transaction->t_reserved_list;
2469                 break;
2470         }
2471 
2472         __blist_add_buffer(list, jh);
2473         jh->b_jlist = jlist;
2474 
2475         if (was_dirty)
2476                 set_buffer_jbddirty(bh);
2477 }
2478 
2479 void jbd2_journal_file_buffer(struct journal_head *jh,
2480                                 transaction_t *transaction, int jlist)
2481 {
2482         jbd_lock_bh_state(jh2bh(jh));
2483         spin_lock(&transaction->t_journal->j_list_lock);
2484         __jbd2_journal_file_buffer(jh, transaction, jlist);
2485         spin_unlock(&transaction->t_journal->j_list_lock);
2486         jbd_unlock_bh_state(jh2bh(jh));
2487 }
2488 
2489 /*
2490  * Remove a buffer from its current buffer list in preparation for
2491  * dropping it from its current transaction entirely.  If the buffer has
2492  * already started to be used by a subsequent transaction, refile the
2493  * buffer on that transaction's metadata list.
2494  *
2495  * Called under j_list_lock
2496  * Called under jbd_lock_bh_state(jh2bh(jh))
2497  *
2498  * jh and bh may be already free when this function returns
2499  */
2500 void __jbd2_journal_refile_buffer(struct journal_head *jh)
2501 {
2502         int was_dirty, jlist;
2503         struct buffer_head *bh = jh2bh(jh);
2504 
2505         J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2506         if (jh->b_transaction)
2507                 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2508 
2509         /* If the buffer is now unused, just drop it. */
2510         if (jh->b_next_transaction == NULL) {
2511                 __jbd2_journal_unfile_buffer(jh);
2512                 return;
2513         }
2514 
2515         /*
2516          * It has been modified by a later transaction: add it to the new
2517          * transaction's metadata list.
2518          */
2519 
2520         was_dirty = test_clear_buffer_jbddirty(bh);
2521         __jbd2_journal_temp_unlink_buffer(jh);
2522         /*
2523          * We set b_transaction here because b_next_transaction will inherit
2524          * our jh reference and thus __jbd2_journal_file_buffer() must not
2525          * take a new one.
2526          */
2527         jh->b_transaction = jh->b_next_transaction;
2528         jh->b_next_transaction = NULL;
2529         if (buffer_freed(bh))
2530                 jlist = BJ_Forget;
2531         else if (jh->b_modified)
2532                 jlist = BJ_Metadata;
2533         else
2534                 jlist = BJ_Reserved;
2535         __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
2536         J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2537 
2538         if (was_dirty)
2539                 set_buffer_jbddirty(bh);
2540 }
2541 
2542 /*
2543  * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2544  * bh reference so that we can safely unlock bh.
2545  *
2546  * The jh and bh may be freed by this call.
2547  */
2548 void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
2549 {
2550         struct buffer_head *bh = jh2bh(jh);
2551 
2552         /* Get reference so that buffer cannot be freed before we unlock it */
2553         get_bh(bh);
2554         jbd_lock_bh_state(bh);
2555         spin_lock(&journal->j_list_lock);
2556         __jbd2_journal_refile_buffer(jh);
2557         jbd_unlock_bh_state(bh);
2558         spin_unlock(&journal->j_list_lock);
2559         __brelse(bh);
2560 }
2561 
2562 /*
2563  * File inode in the inode list of the handle's transaction
2564  */
2565 static int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode,
2566                                    unsigned long flags)
2567 {
2568         transaction_t *transaction = handle->h_transaction;
2569         journal_t *journal;
2570 
2571         if (is_handle_aborted(handle))
2572                 return -EROFS;
2573         journal = transaction->t_journal;
2574 
2575         jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2576                         transaction->t_tid);
2577 
2578         /*
2579          * First check whether inode isn't already on the transaction's
2580          * lists without taking the lock. Note that this check is safe
2581          * without the lock as we cannot race with somebody removing inode
2582          * from the transaction. The reason is that we remove inode from the
2583          * transaction only in journal_release_jbd_inode() and when we commit
2584          * the transaction. We are guarded from the first case by holding
2585          * a reference to the inode. We are safe against the second case
2586          * because if jinode->i_transaction == transaction, commit code
2587          * cannot touch the transaction because we hold reference to it,
2588          * and if jinode->i_next_transaction == transaction, commit code
2589          * will only file the inode where we want it.
2590          */
2591         if ((jinode->i_transaction == transaction ||
2592             jinode->i_next_transaction == transaction) &&
2593             (jinode->i_flags & flags) == flags)
2594                 return 0;
2595 
2596         spin_lock(&journal->j_list_lock);
2597         jinode->i_flags |= flags;
2598         /* Is inode already attached where we need it? */
2599         if (jinode->i_transaction == transaction ||
2600             jinode->i_next_transaction == transaction)
2601                 goto done;
2602 
2603         /*
2604          * We only ever set this variable to 1 so the test is safe. Since
2605          * t_need_data_flush is likely to be set, we do the test to save some
2606          * cacheline bouncing
2607          */
2608         if (!transaction->t_need_data_flush)
2609                 transaction->t_need_data_flush = 1;
2610         /* On some different transaction's list - should be
2611          * the committing one */
2612         if (jinode->i_transaction) {
2613                 J_ASSERT(jinode->i_next_transaction == NULL);
2614                 J_ASSERT(jinode->i_transaction ==
2615                                         journal->j_committing_transaction);
2616                 jinode->i_next_transaction = transaction;
2617                 goto done;
2618         }
2619         /* Not on any transaction list... */
2620         J_ASSERT(!jinode->i_next_transaction);
2621         jinode->i_transaction = transaction;
2622         list_add(&jinode->i_list, &transaction->t_inode_list);
2623 done:
2624         spin_unlock(&journal->j_list_lock);
2625 
2626         return 0;
2627 }
2628 
2629 int jbd2_journal_inode_add_write(handle_t *handle, struct jbd2_inode *jinode)
2630 {
2631         return jbd2_journal_file_inode(handle, jinode,
2632                                        JI_WRITE_DATA | JI_WAIT_DATA);
2633 }
2634 
2635 int jbd2_journal_inode_add_wait(handle_t *handle, struct jbd2_inode *jinode)
2636 {
2637         return jbd2_journal_file_inode(handle, jinode, JI_WAIT_DATA);
2638 }
2639 
2640 /*
2641  * File truncate and transaction commit interact with each other in a
2642  * non-trivial way.  If a transaction writing data block A is
2643  * committing, we cannot discard the data by truncate until we have
2644  * written them.  Otherwise if we crashed after the transaction with
2645  * write has committed but before the transaction with truncate has
2646  * committed, we could see stale data in block A.  This function is a
2647  * helper to solve this problem.  It starts writeout of the truncated
2648  * part in case it is in the committing transaction.
2649  *
2650  * Filesystem code must call this function when inode is journaled in
2651  * ordered mode before truncation happens and after the inode has been
2652  * placed on orphan list with the new inode size. The second condition
2653  * avoids the race that someone writes new data and we start
2654  * committing the transaction after this function has been called but
2655  * before a transaction for truncate is started (and furthermore it
2656  * allows us to optimize the case where the addition to orphan list
2657  * happens in the same transaction as write --- we don't have to write
2658  * any data in such case).
2659  */
2660 int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2661                                         struct jbd2_inode *jinode,
2662                                         loff_t new_size)
2663 {
2664         transaction_t *inode_trans, *commit_trans;
2665         int ret = 0;
2666 
2667         /* This is a quick check to avoid locking if not necessary */
2668         if (!jinode->i_transaction)
2669                 goto out;
2670         /* Locks are here just to force reading of recent values, it is
2671          * enough that the transaction was not committing before we started
2672          * a transaction adding the inode to orphan list */
2673         read_lock(&journal->j_state_lock);
2674         commit_trans = journal->j_committing_transaction;
2675         read_unlock(&journal->j_state_lock);
2676         spin_lock(&journal->j_list_lock);
2677         inode_trans = jinode->i_transaction;
2678         spin_unlock(&journal->j_list_lock);
2679         if (inode_trans == commit_trans) {
2680                 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
2681                         new_size, LLONG_MAX);
2682                 if (ret)
2683                         jbd2_journal_abort(journal, ret);
2684         }
2685 out:
2686         return ret;
2687 }
2688 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp